Cognition Binaries: Contrasts in Mind & AI

• Cognition binaries are a set of recurring distinctions (e.g., cognizer/tool, mental/non-mental) that demarcate debates on what constitutes cognition across various fields.
• They are used to contrast simulation versus implementation, dual-process models, and binary versus graded representations, impacting both theoretical models and system designs.
• These distinctions offer practical insights into operationalizing cognition, influencing methodologies in AI, cognitive science, and philosophy of mind.

Cognition binaries designates a family of distinctions used to delimit, model, or operationalize cognition across philosophy of mind, cognitive science, artificial intelligence, formal language theory, and human–AI systems. In the most explicit formulation, the term names contrasts such as cognizer versus tool, mental versus non-mental, narrow versus wide cognition, and individual mind versus distributed systems; in adjacent literatures it also refers to simulation versus implementation, intuition versus deliberation, active versus inactive representational units, and finite-state versus stack-based computational architectures (0808.3569, Harnad, 2012, Reimann, 2021, Granger, 2020). Taken together, these works suggest that “cognition binaries” is not a single doctrine but a recurrent way of organizing disputes about what cognition is, where it resides, how it is implemented, and whether its structure is best understood as categorical or graded.

1. Conceptual range and recurrent distinctions

The literature groups cognition binaries into a small number of recurring contrast classes. Some are ontological, some computational, and some methodological. In the philosophical strand, the decisive contrast is between systems with mental states and systems that merely contribute functionally to cognition. In the computational strand, the decisive contrast is often between binary-coded internal states and richer structures built from them. In the social and behavioral strand, the contrast is frequently between intuitive and deliberative processing. More recent work argues that these oppositions are often too rigid and should be replaced by multidimensional cognition spaces (0808.3569, Signorelli, 2017, Reimann, 2021, Solé et al., 19 Jan 2026).

Strand	Characteristic binaries	Representative papers
Philosophy of mind	cognizer/tool; mental/non-mental; narrow/wide; individual/distributed	(0808.3569)
Causal explanation	simulation/implementation; computational/dynamical; visible/invisible	(Harnad, 2012)
Formal representation	on/off units; active/inactive patterns; binding/bundling states	(Cho et al., 2022, Reimann, 2021, Yang, 2023)
Dual-process and social cognition	intuition/deliberation; conscious/non-conscious; assortative/non-assortative matching	(Signorelli, 2017, Bilancini et al., 2022)
Post-binary frameworks	graded cognition spaces; natural/artificial/hybrid continua	(Solé et al., 19 Jan 2026)

This range matters because identical vocabulary can support opposite conclusions. In one tradition, offloading and distributed performance are used to defend sharp boundaries; in another, binary encodings are treated as the substrate from which compositional cognition can be built; in a third, binary distinctions are criticized as projections of a richer landscape (0808.3569, Chuma et al., 20 May 2026, Solé et al., 19 Jan 2026).

2. Mind, tool, and the defense of sharp boundaries

The strongest defense of cognition binaries is the claim that only mental states are cognitive states. On this view, a mental state is a felt state or conscious state; a cognizer is a system that has such states; cognition is mental states together with the bodily and brain processes that implement them. Systems without mental states may function as cognitive technology, but they do not become cognizers merely by contributing to human performance (0808.3569).

This position preserves several linked binaries. The primary distinction is mental versus non-mental. A secondary distinction separates cognitive from non-cognitive among mental processes. From this perspective, “narrow cognition” is “skin and in”: the internal mechanism that implements thinking, understanding, remembering, and related conscious states. External memory aids, notebooks, databases, search engines, and the web can expand performance, but they remain inputs and outputs of cognition rather than parts of the mental state itself. The paper’s “migraine test” is decisive here: a migraine cannot be distributed over Congress or the US government, and no coherent sense exists in which a group or institution literally has a headache. The same argument is extended to group cognition: collaborative cognition is real, but a team is not a superordinate mind (0808.3569).

Language occupies a special position because it is simultaneously a cognitive capacity and a cognitive technology. It allows one cognizer to offload memory, reasoning, and problem-solving onto other cognizers, and later media such as writing, print, telecommunications, computers, and the web intensify this process. The web is described as a “Cognitive Commons,” a distributed network of cognizers, digital databases, and software agents that greatly increases cognitive performance. Yet the ontological boundary remains: the Commons is a shared environment of tools and other minds, not a single cognizer (0808.3569).

3. Simulation, implementation, and causal topography

A second major family of cognition binaries concerns causal explanation. Here the central distinction is between a system that simulates cognition and a system that implements it. Computation is defined as syntactic symbol manipulation, implementation-independent and organizational. A computation can formally mirror a target process, but formal mirroring does not by itself confer the causal powers of the target (Harnad, 2012).

The contrast is expressed with two relations. If $S(M,P)$, model $M$ simulates process $P$; if $I(M,P)$, system $M$ implements process $P$. The argument is that cognition resembles heating, flying, or planetary motion in this respect: a computation can simulate their causal structure, but it does not instantiate the relevant dynamical properties. A computational furnace does not heat a house; a planetary simulator does not literally orbit; likewise, a computational model of cognition does not thereby feel or cognize. The temptation to infer otherwise arises because feeling is invisible to everyone except the cognizer. What is third-person available is behavior and brain activity; what is first-person only is experience itself (Harnad, 2012).

This yields several related binaries: computational versus dynamical, organizational versus physical, observable versus invisible, behavioral capacity versus conscious experience. The paper accepts a full robotic Turing Test as a test of doing, not of feeling. It therefore endorses computational explanation while rejecting computational sufficiency. The causal topography of cognition is said to involve a specific mix of dynamics—sensorimotor, neurobiological, molecular, and perhaps computational—and the paper names “dynamism” as the thesis that some non-computational dynamics are essential to cognizing and feeling (Harnad, 2012).

4. Binary representations, neural codes, and algebraic cognition

A different use of cognition binaries appears in representational and architectural work that models cognition directly with binary-like states. In the neural-lexicon hypothesis, cognition is organized as a layered pathway from Sensogram to Engram to Pre-Langram to Langram to Cognogram to Decigram and finally Circulogram, with thinking operationally identified with the Circulogram’s state at $t=t_n$ in the absence of new sensory input. The model can be read as a hierarchy of quasi-discrete selections: neurons are effectively on or off, Engram membership is included or excluded, and stored patterns are resonant or non-resonant. At each stage, some units are kept and others discarded (Cho et al., 2022).

Working-memory modeling pushes this further by representing a cognitive state as a high-dimensional binary vector in a Hamming-like space. Binding is a reversible component-wise operation, and bundling is a stochastic addition $+_p$ that is explicitly non-associative for $0Reimann, 2021).

Hyperdimensional approaches generalize the same idea. Sparse binary hypervectors treat concepts, roles, words, and sequences as long binary patterns, with bundle and bind operations supporting memory, role–filler composition, sets, sequences, and analogy. Online bundling yields a streaming learner $L(k)$ that updates during inference, so training and use are not sharply separated. A related hardware-oriented line proposes a semi-holographic representation using only multiplexing and addition operations, with a “cognitive processing unit” that performs superposition and binding for 64-bit operands below 6 pJ. Another recent architecture, organized around XOR-and-shift over $M$0, argues that reversible variable binding, recursively structured representations, and the distinction between individuals and kinds can be realized by a single algebraic primitive implemented with primitive-polynomial linear-feedback shift registers (Yang, 2023, Serb et al., 2019, Chuma et al., 20 May 2026).

5. Dual-process binaries, categorization, and social matching

In dual-process work, cognition binaries take the form of paired processing modes. A four-type taxonomy distinguishes contents from processes and then classifies each as conscious or non-conscious, yielding Type 0, Type 1, Type 2, and the speculative Type o. Type 0 is non-conscious content with non-conscious processing and is described as “computation-light but learning-heavy.” Type 2 is conscious content with deliberate, controlled processing and is described as “computation-heavy, learning-light and interferes with Type 0.” The paper uses this taxonomy to argue that higher-level cognitive machines would trade accuracy, speed, and obedience for consciousness and autonomy (Signorelli, 2017).

Experimental work on binary categorization examines what happens when cognition is forced into two-category judgments. Across gray-color classification, vowel categorization, and number discrimination, psychometric tails and response times are interpreted as evidence for a transition between two regimes: conscious choice of the non-dominant category in the uncertainty region and physical errors in obvious cases. The proposed explanation is dual-system decision making: System 2 dominates under uncertainty and yields longer response times, whereas System 1 dominates when the category is obvious and residual errors flatten into a motor or recording floor (Lubashevsky et al., 2019).

A social extension treats intuition and deliberation as binary cognitive modes that can assort in pairwise interaction. Assortativity in cognition is defined by $M$1, equivalently $M$2. In the applied model, increasing assortativity raises the probability of same-mode pairs and can increase cooperation by altering how intuitive heuristics are shaped through interaction and reinforcement. Yet the welfare effects are not monotone across all games: depending on the payoff structure, cognitive assortativity can be socially desirable or undesirable (Bilancini et al., 2022).

Not all dual-process work accepts the binary architecture. A neurosymbolic alternative argues that cognition is not fundamentally System-1/System-2 in this way, because both low- and high-level cognition contain symbolic and subsymbolic information. On that view, the main differentiating factor is attention, not a hard split between symbolic/high-level and subsymbolic/low-level processing (Latapie et al., 2021).

6. Formal limits, logical agents, and the quantification of cognitive power

Another strand treats cognition binaries as formal capability boundaries. One proposal decomposes every cognitive system into a finite-state machine plus a memory architecture, then asks what memory operations are available. On this account, a pure FSM recognizes regular languages; a single pushdown stack yields context-free power; nested stacks yield indexed or mildly context-sensitive power; multiple bounded stacks yield context-sensitive power; and multiple unbounded stacks yield Turing equivalence. Intersecting human cognitive abilities, allometric brain structure, and automata theory, the paper argues that human intrinsic cognition is best characterized as a nested-stack architecture rather than a Turing machine (Granger, 2020).

A related normative analysis derives a Requirement Equation for cognition: $M$3 The paper uses this to distinguish two regimes. In low-information action spaces, shortcut cognition can approximate the requirement with simple mappings from sensory data to action. In complex domains such as 3D spatial cognition, where many actions depend on the same hidden structure, only Bayesian internal models can satisfy the requirement well. This creates a formal binary between shortcut and model-based cognition, although the paper presents it as a threshold-like shift rather than an absolute metaphysical divide (Worden, 2024).

A further logical formalization appears in Cognitive Binary Logic. Here the agent’s discourse is built around assertions $M$4 and queries $M$5, and the dialog function maps inputs into the pair logical/nonsense together with tautology, contradiction, or contextual truth. The resulting Cognitive Intelligent Agent is designed to pass a family of Turing Tests concerning the modal truth states of formulas in propositional binary logic without error. In this formulation, cognition binaries are literal logical distinctions: assertion versus question, logical versus nonsense, and necessary truth versus contradiction, with contextual truth as the residual case (Popescu-Bodorin et al., 2011).

7. From binary oppositions to graded cognition spaces

Recent work explicitly rejects the idea that cognition is best understood through sharp binaries. The cognition-space approach defines cognition broadly as the capacity to sense, process, and respond to information, then maps natural, artificial, and hybrid systems into multidimensional morphospaces. Three such spaces are proposed: a basal aneural space organized by spatial complexity, computational complexity, and developmental complexity; a neural space organized by individual agency, computational complexity, and agent–agent interaction complexity; and a human–AI hybrid space organized by human cognition, AI cognition, and human–AI exchange (Solé et al., 19 Jan 2026).

The important result is that occupation of these spaces is highly uneven. There are clusters of realized systems and large unoccupied regions. Those voids are interpreted not as categorical proof of non-cognition, but as products of evolutionary contingency, physical constraint, and design limitation. Hybrid systems—organoids in microfluidic chips, robots embedded in animal societies, human–AI dyads, and “humanbots”—are especially significant because they occupy regions that dissolve older oppositions such as natural versus artificial, individual versus collective, and human versus machine (Solé et al., 19 Jan 2026).

An applied version of this post-binary move appears in AI-driven education. The Expert Cognition Dashboard argues that behavior alone cannot fully represent cognition and replaces the behavior/cognition binary with a pipeline in which learner behaviors are transformed by AI Tutor analysis into interpretable cognition structures, then aggregated across individual, class, and AI Twin dashboards. In that framework, dashboards become cognitive middleware connecting behavioral traces to expert-like AI reasoning rather than merely displaying performance metrics (Yuan, 17 May 2026).

Cognition binaries therefore name two opposed tendencies in contemporary theory. One tendency uses binary distinctions to preserve conceptual clarity about mind, implementation, and agency. The other uses binary states, codes, and algebraic operations as constructive substrates for modeling cognition. A third, increasingly prominent tendency argues that both projects remain incomplete unless binaries are embedded in graded spaces, where differences in cognition are analyzed as differences of organization, complexity, agency, and coupling rather than as exhaustive either/or kinds (0808.3569, Solé et al., 19 Jan 2026).